{"paper_id":"4453cda5-e0ec-4325-b788-1d2454b1c38d","body_text":"Preoperative Prediction of Histological Grade in Pancreatic Neuroendocrine Tumors Using Whole-Pancreas CT Radiomics and Clinical Features: A Single-Center Retrospective Study | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Preoperative Prediction of Histological Grade in Pancreatic Neuroendocrine Tumors Using Whole-Pancreas CT Radiomics and Clinical Features: A Single-Center Retrospective Study Ximing Liu, Yujie Zhou, Shuang Gao, Wenpei Zhang, Rixiong Wang This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9427497/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 6 You are reading this latest preprint version Abstract Objective To develop and internally validate a whole-pancreas CT radiomics model, a clinical model, and a combined model for the preoperative prediction of histological grade in pancreatic neuroendocrine tumors (pNETs). Methods This single-center retrospective study included 79 pathologically confirmed pNETs between January 2019 and December 2024. The endpoint was binary histological grade (G1 vs G2/G3). Whole-pancreas segmentation was performed on preoperative multiphasic CT images, and radiomics signatures were generated within each training fold using standardization, near-zero variance filtering, redundancy reduction, and L1-penalized feature selection. Four clinical variables (Ki-67, maximum tumor diameter, T stage, and liver metastasis) were selected based on univariate analysis and clinical relevance. Three logistic models were developed: a clinical model, a radiomics model, and a combined model. Internal validation used nested repeated stratified 5-fold cross-validation (20 repeats in the outer loop and 3 folds in the inner loop) with 1000 bootstrap resamples to estimate 95% confidence intervals (CIs) for performance metrics. Results Of the 79 patients, 31 had G1 tumors and 48 had G2/G3 tumors. The combined model showed the best overall performance, with an AUC of 0.798, accuracy of 0.759, sensitivity of 0.771, specificity of 0.742, F1 score of 0.796, and Brier score of 0.182. The clinical and radiomics models yielded AUCs of 0.780 and 0.776, respectively. Bootstrap comparison showed no statistically stable AUC difference between the combined and clinical models (delta AUC 0.015, 95% CI -0.116 to 0.138) or between the combined and radiomics models (delta AUC 0.021, 95% CI -0.037 to 0.065). Conclusion A whole-pancreas CT radiomics-clinical combined model showed the most balanced internal validation performance for preoperative grading of pNETs, although its incremental benefit over the single-domain models remains uncertain in this small cohort. Larger multicenter studies with external validation are needed. pancreatic neuroendocrine tumors CT radiomics whole-pancreas segmentation histological grade combined model bootstrap validation Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 1. Introduction Pancreatic neuroendocrine tumors (pNETs) are a heterogeneous group of neoplasms with markedly different biological behaviors and clinical outcomes[1]. Histological grade is one of the most important determinants of prognosis and treatment selection. In clinical practice, low-grade tumors may be considered for surveillance or limited surgery in selected circumstances, whereas higher-grade tumors generally require a more aggressive therapeutic strategy[2,3]. Accurate preoperative estimation of tumor grade is therefore central to personalized management. At present, preoperative grading relies mainly on endoscopic ultrasound-guided fine-needle aspiration or biopsy[4,5]. Although this approach remains important, it is invasive and may be affected by sampling error because of intratumoral heterogeneity[5]. Conventional contrast-enhanced CT can depict tumor size, location, enhancement pattern, and metastatic status, but visual assessment alone is often insufficient for reliable estimation of histological grade[6,7]. Radiomics provides a noninvasive approach for mining quantitative image information that may capture latent tumor heterogeneity[8]. Most previous studies focused on manually outlined tumor regions only. In contrast, whole-pancreas segmentation may additionally reflect subtle changes in the surrounding pancreatic parenchyma and tumor-associated microenvironment[9,10]. The present study therefore aimed to construct and compare three preoperative grading strategies based on whole-pancreas CT radiomics, clinical variables, and their combination, using a repeated nested cross-validation framework to reduce optimistic bias and to provide a more robust estimate of internal performance[11]. 2. Materials and Methods 2.1 Study design and patient population This was a retrospective single-center study that included consecutive patients with pathologically confirmed pNETs treated at the First Affiliated Hospital of Fujian Medical University between January 2019 and December 2024. Eligible patients had available preoperative CT images and definitive postoperative pathological grading. Patients with incomplete imaging data, unavailable pathology, or prior treatment before CT were excluded. The primary study endpoint was histological grade recoded as a binary outcome: G1 tumors were defined as the low-grade group and G2/G3 tumors were combined as the high-grade group. No fixed training/testing split was used. Instead, all patients contributed to model development and internal validation through repeated nested cross-validation. 2.2 CT imaging, whole-pancreas segmentation, and radiomics workflow CT examinations were acquired on multidetector scanners using routine institutional protocols. Whole-pancreas segmentation was performed manually using ITK-SNAP by a radiologist with 3 years of experience and reviewed by a senior radiologist with 20 years of experience. In a subset of randomly selected cases, repeated segmentation was used to assess reproducibility, and only stable features were retained for downstream analysis. Radiomics features were extracted from multiphasic CT images using PyRadiomics (version 3.0). Within each training fold, features underwent z-score standardization, near-zero variance filtering, correlation-based redundancy reduction, and L1-penalized logistic feature selection. The selected features were combined into a fold-specific radiomics score (Rad-score)[12]. 2.3 Clinical variable selection and model development Candidate clinical variables were screened using univariate analysis and clinical interpretability. Four variables were ultimately retained for the clinical model: Ki-67, maximum tumor diameter, T stage, and liver metastasis[13,14]. Three logistic regression models were then built: (1) a clinical model using the four clinical predictors, (2) a radiomics model using the fold-specific Rad-score, and (3) a combined model integrating the clinical variables with the Rad-score. Internal validation used repeated stratified 5-fold cross-validation in the outer loop (20 repeats, yielding 100 outer-fold evaluations) and stratified 3-fold cross-validation in the inner loop for hyperparameter tuning and feature selection. This design ensured that all preprocessing and feature selection steps were repeated strictly within the training data to avoid information leakage. 2.4 Model evaluation and statistical analysis Model discrimination and classification performance were summarized using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, F1 score, and Brier score. Out-of-fold predictions from repeated cross-validation were used to generate ROC and calibration plots. Bootstrap resampling (1000 iterations) was performed to estimate 95% confidence intervals for each performance metric and for pairwise differences in AUC. Continuous variables are presented as median (interquartile range) and categorical variables as counts (percentages). Group comparisons were performed with the Mann-Whitney U test, chi-square test, or Fisher exact test as appropriate. All tests were two-sided, and P values < 0.05 were considered statistically significant. 3. Results 3.1 Patient characteristics A total of 79 patients were included, including 31 with G1 tumors and 48 with G2/G3 tumors. Baseline comparisons between the two grade groups are summarized in Table 1 . Patients in the high-grade group had larger tumors (median 26.8 mm vs 19.0 mm, P = 0.002), higher Ki-67 values (median 5.0% vs 1.0%, P < 0.001), and more advanced T stage (P = 0.019). Age and liver metastasis did not differ significantly between the two groups. Table 1 Baseline characteristics according to histological grade group. Variable G1 (n = 31) G2/G3 (n = 48) P value Age at diagnosis, years 58.0 (54.0-69.5) 57.5 (46.5–68.0) 0.154 Maximum tumor diameter, mm 19.0 (15.0-25.3) 26.8 (20.0-40.4) 0.002 Ki-67, % 1.0 (1.0–2.0) 5.0 (3.4–13.9) < 0.001 T stage 0.019 T1 16 (51.6%) 11 (22.9%) T2 12 (38.7%) 24 (50.0%) T3 3 (9.7%) 13 (27.1%) Liver metastasis, yes 3 (9.7%) 12 (25.0%) 0.141 Data are presented as median (interquartile range) or n (%). P values were derived using the Mann-Whitney U test, chi-square test, or Fisher exact test, as appropriate. 3.2 Internal validation performance of the three models Internal validation performance is summarized in Table 2 . The combined model demonstrated the best overall balance of discrimination and calibration, with an AUC of 0.798, accuracy of 0.759, sensitivity of 0.771, specificity of 0.742, F1 score of 0.796, and Brier score of 0.182. The clinical model provided slightly higher specificity (0.806), whereas the radiomics-only model achieved the highest specificity (0.903) but lower sensitivity (0.563). ROC and calibration plots are shown in Figs. 1 and 2 , and the bootstrap-based comparison across metrics is shown in Fig. 3 . Table 2 Bootstrap-based internal validation performance of the three prediction models. Metric Clinical model Radiomics model Combined model AUC 0.780 (0.676–0.878) 0.776 (0.660–0.881) 0.798 (0.680–0.897) Accuracy 0.696 (0.595–0.797) 0.696 (0.595–0.797) 0.759 (0.658–0.848) Sensitivity 0.625 (0.479–0.756) 0.562 (0.413–0.714) 0.771 (0.636–0.886) Specificity 0.806 (0.656–0.935) 0.903 (0.783-1.000) 0.742 (0.576–0.893) F1 score 0.714 (0.585–0.813) 0.692 (0.557–0.804) 0.796 (0.690–0.880) Brier score 0.214 (0.188–0.242) 0.237 (0.228–0.246) 0.182 (0.130–0.247) Values are point estimates with bootstrap 95% confidence intervals in parentheses. The combined model had the lowest Brier score, indicating the best overall calibration. 3.3 Pairwise AUC comparison and radiomics feature stability Despite the favorable point estimates of the combined model, bootstrap analysis of pairwise AUC differences did not demonstrate statistically stable superiority (Table 3 and Fig. 4 ). The delta AUC was 0.015 for combined minus clinical and 0.021 for combined minus radiomics, and both confidence intervals crossed zero. Repeated training runs showed that radiomics features related to size-zone non-uniformity, dependence variance, size, and first-order intensity were most frequently selected (Fig. 5 ). In the final combined model fitted on the full dataset, the coefficient of the Rad-score (9.640) was larger than those of the individual clinical variables, suggesting that the aggregated radiomics signature remained an important contributor to prediction. Table 3 Pairwise differences in AUC estimated by 1000 bootstrap resamples. Comparison Delta AUC 95% CI Combined minus clinical 0.015 -0.116 to 0.138 Combined minus radiomics 0.021 -0.037 to 0.065 Clinical minus radiomics 0.006 -0.133 to 0.155 Confidence intervals crossing zero indicate that a stable difference in discrimination was not established. 4. Discussion This study developed a whole-pancreas CT radiomics strategy for preoperative prediction of pNET histological grade and compared it with a clinical-only model and a combined model. The main finding was that the combined model showed the most balanced internal validation profile, with the highest AUC and F1 score as well as the lowest Brier score. However, the estimated performance gain over the single-domain models was modest, and bootstrap confidence intervals for the AUC differences crossed zero. The results therefore support cautious interpretation rather than a claim of clear superiority[15,16]. An important feature of the present work is the use of whole-pancreas segmentation rather than tumor-only segmentation. Whole-organ analysis may capture subtle alterations in the pancreatic background parenchyma, vascular environment, or tumor-associated heterogeneity that are not fully represented by a focal tumor ROI[17]. The frequent selection of radiomics features associated with heterogeneity and size-zone distribution also supports the biological plausibility of the radiomics signature[18,19]. The current modeling strategy was intentionally conservative. Instead of relying on a single random training/testing split, we used repeated nested stratified cross-validation together with bootstrap-based uncertainty estimation[20]. This approach is better suited to small datasets and helps reduce optimistic bias. The trade-off is that any apparent advantage of one model over another must remain large and stable to survive resampling-based comparison[21,22]. In the present cohort, the combined model appeared promising but not definitively superior. Several limitations should be acknowledged. First, this was a single-center retrospective study with a relatively small sample size. Second, no external validation cohort was available, and therefore generalizability remains uncertain. Third, although the combined model incorporated key clinical variables and radiomics information, other potentially relevant biomarkers were not evaluated in the current framework. Future multicenter studies should test the transportability of the model, refine feature harmonization across scanners, and explore integration with additional imaging or molecular markers. 5. Conclusion Whole-pancreas CT radiomics combined with clinical variables provided the best overall internal validation performance for the preoperative prediction of pNET histological grade in this cohort. Nevertheless, the incremental discrimination gain over the clinical and radiomics-only models was small and not statistically stable in bootstrap comparison. External validation in larger cohorts is required before clinical implementation. Declarations Ethics approval: Ethical approval (2023J0112) was obtained from the ethics review board of The Fujian Medical University; the requirement for informed consent was waived owing to the retrospective nature of the study. Conflict of interest: The authors declare no competing interests. Funding: This work was supported by 1. Fujian Provincial Natural Science Foundation (2023J0112); 2. Fujian Provincial Financial Special Fund (BRB-2024WRX) 3. Wu Jieping Medical Foundation(320.6750.2023-05-116). Author Contribution Author contributionsXiming Liu contributed to study conception and design, collected the clinical and imaging data, and drafted the manuscript. Yujie Zhou contributed to data analysis, model development, and manuscript revision. Shuang Gao contributed to statistical analysis and revision of the manuscript. Wenpei Zhang contributed to data curation, methodological discussion, and manuscript editing. Rixiong Wang conceived and supervised the study, interpreted the results, and critically revised the manuscript for important intellectual content. All authors reviewed and approved the final manuscript. Unsectioned Paragraphs Ma, Z.-Y. et al. Pancreatic neuroendocrine tumors: A review of serum biomarkers, staging, and management. World J. Gastroenterol. 26 , 2305–2322 (2020). N F. et al. Natural history and clinical outcomes of pancreatic neuroendocrine neoplasms based on the WHO 2017 classification; a single-center experience of 30 years. Pancreatol. Off. J. Int. Assoc. Pancreatol. IAP Al 20 , (2020). Pavel, M. et al. Gastroenteropancreatic neuroendocrine neoplasms: ESMO Clinical Practice Guidelines for diagnosis, treatment and follow-up. Ann. Oncol. Off. J. Eur. Soc. Med. Oncol. 31 , 844–860 (2020). Ardengh, J. C. et al. Pancreatic splenosis mimicking neuroendocrine tumors: microhistological diagnosis by endoscopic ultrasound guided fine needle aspiration. Arq. Gastroenterol. 50 , 10–14 (2013). Ye, X. et al. Endoscopic Ultrasound-Guided Fine Needle Acquisition for Evaluation of Pancreatic Neuroendocrine Tumors: A Meta-Analysis. J. Clin. Gastroenterol. 59 , 310–320 (2025). Lee, L., Ito, T. & Jensen, R. T. Imaging of pancreatic neuroendocrine tumors: recent advances, current status, and controversies. Expert Rev. Anticancer Ther. 18 , 837–860 (2018). Tacelli, M. et al. Reliability of grading preoperative pancreatic neuroendocrine tumors on EUS specimens: a systematic review with meta-analysis of aggregate and individual data. Gastrointest. Endosc. 96 , 898-908.e23 (2022). Lambin, P. et al. Radiomics: the bridge between medical imaging and personalized medicine. Nat. Rev. Clin. Oncol. 14 , 749–762 (2017). Kumar, V. et al. Radiomics: the process and the challenges. Magn. Reson. Imaging 30 , 1234–1248 (2012). Wang, S. et al. Compute Tomography Radiomics Analysis on Whole Pancreas Between Healthy Individual and Pancreatic Ductal Adenocarcinoma Patients: Uncertainty Analysis and Predictive Modeling. Technol. Cancer Res. Treat. 21 , 15330338221126869 (2022). W G. et al. Development and validation of CT-based radiomics deep learning signatures to predict lymph node metastasis in non-functional pancreatic neuroendocrine tumors: a multicohort study. EClinicalMedicine 65 , (2023). Shi, Z., Traverso, A., van Soest, J., Dekker, A. & Wee, L. Technical Note: Ontology-guided radiomics analysis workflow (O-RAW). Med. Phys. 46 , 5677–5684 (2019). F, G. et al. KI-67 heterogeneity in well differentiated gastro-entero-pancreatic neuroendocrine tumors: when is biopsy reliable for grade assessment? Endocrine 57 , (2017). Development and validation of a prognostic nomogram based on inflammation and Nutrition-Related Indexes for predicting postoperative survival in patients with pancreatic neuroendocrine tumors - PubMed. https://pubmed.ncbi.nlm.nih.gov/41469599/. Snell, K. I. E. et al. Transparent reporting of multivariable prediction models for individual prognosis or diagnosis: checklist for systematic reviews and meta-analyses (TRIPOD-SRMA). BMJ 381 , e073538 (2023). Zwanenburg, A. et al. The Image Biomarker Standardization Initiative: Standardized Quantitative Radiomics for High-Throughput Image-based Phenotyping. Radiology 295 , 328–338 (2020). Kang, W. et al. Application of radiomics-based multiomics combinations in the tumor microenvironment and cancer prognosis. J. Transl. Med. 21 , 598 (2023). Lambin, P. et al. Radiomics: extracting more information from medical images using advanced feature analysis. Eur. J. Cancer 48 , 441–446 (2012). Dou, T. H., Coroller, T. P., van Griethuysen, J. J. M., Mak, R. H. & Aerts, H. J. W. L. Peritumoral radiomics features predict distant metastasis in locally advanced NSCLC. PloS One 13 , e0206108 (2018). Schroeder, A. W., Tran, Z., Sexton, K. & Salzberg, A. D. Clinician’s Guide to Artificial Intelligence: Technical Foundations of Machine Learning. Med. Clin. North Am. 110 , 287–305 (2026). TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods - PubMed. https://pubmed.ncbi.nlm.nih.gov/38626948/. Snell, K. I. E. et al. Transparent reporting of multivariable prediction models for individual prognosis or diagnosis: checklist for systematic reviews and meta-analyses (TRIPOD-SRMA). BMJ 381 , e073538 (2023). Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviewers agreed at journal 07 May, 2026 Reviewers agreed at journal 06 May, 2026 Reviewers invited by journal 16 Apr, 2026 Editor assigned by journal 16 Apr, 2026 Submission checks completed at journal 16 Apr, 2026 First submitted to journal 15 Apr, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {\"props\":{\"pageProps\":{\"initialData\":{\"identity\":\"rs-9427497\",\"acceptedTermsAndConditions\":true,\"allowDirectSubmit\":false,\"archivedVersions\":[],\"articleType\":\"Research Article\",\"associatedPublications\":[],\"authors\":[{\"id\":627550160,\"identity\":\"ba9dbde5-09d3-4483-963e-6dac08900c73\",\"order_by\":0,\"name\":\"Ximing Liu\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"The First Affiliated Hospital of Fujian Medical University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Ximing\",\"middleName\":\"\",\"lastName\":\"Liu\",\"suffix\":\"\"},{\"id\":627550161,\"identity\":\"3d4f8aa4-6f75-4972-b143-7f9b00d6da7e\",\"order_by\":1,\"name\":\"Yujie Zhou\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"The First Affiliated Hospital of Fujian Medical University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Yujie\",\"middleName\":\"\",\"lastName\":\"Zhou\",\"suffix\":\"\"},{\"id\":627550162,\"identity\":\"06b5b3f1-93ca-4eba-bbcf-ddf70d3a0066\",\"order_by\":2,\"name\":\"Shuang Gao\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Huaihe Hospital of Henan University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Shuang\",\"middleName\":\"\",\"lastName\":\"Gao\",\"suffix\":\"\"},{\"id\":627550163,\"identity\":\"cfc0e82f-b930-404a-9b17-fe4e9e063432\",\"order_by\":3,\"name\":\"Wenpei Zhang\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Huaihe Hospital of Henan University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Wenpei\",\"middleName\":\"\",\"lastName\":\"Zhang\",\"suffix\":\"\"},{\"id\":627550164,\"identity\":\"6101f024-7830-479f-b125-ec3f9cbd8caf\",\"order_by\":4,\"name\":\"Rixiong Wang\",\"email\":\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA6ElEQVRIiWNgGAWjYBACNv7mgw8S//2Tk5c///FBQkUNYS18EseSDT6wHTA2nMFgbPDgzDHCWuQYcswEZ7AdSGy4wWAm+bCFmQiHMZwxY+bhuWPMOLshrSKxgY2Bv707Ab8W5rayxzwSz+TYZQ4cu5G4Q4ZB4szZDQRsObzdmMeA2ZixIbHtRuIZNgYDiVxCWhLMpHkSmBMbDiSzFSS2MROjJcVMcsaBw0Dvp7ExEKcFFMgfG9KMDXvOMEsknDnGQ9Av8v2gqGywkZNn72H8+KOiRo6/vRe/FgzAQ5ryUTAKRsEoGAVYAQB74EyZhw4sLQAAAABJRU5ErkJggg==\",\"orcid\":\"\",\"institution\":\"The First Affiliated Hospital of Fujian Medical University\",\"correspondingAuthor\":true,\"prefix\":\"\",\"firstName\":\"Rixiong\",\"middleName\":\"\",\"lastName\":\"Wang\",\"suffix\":\"\"}],\"badges\":[],\"createdAt\":\"2026-04-15 13:10:44\",\"currentVersionCode\":1,\"declarations\":\"\",\"doi\":\"10.21203/rs.3.rs-9427497/v1\",\"doiUrl\":\"https://doi.org/10.21203/rs.3.rs-9427497/v1\",\"draftVersion\":[],\"editorialEvents\":[],\"editorialNote\":\"\",\"failedWorkflow\":false,\"files\":[{\"id\":107834645,\"identity\":\"39ac7a1f-b049-4f4d-b9dd-f9b2d517aa9f\",\"added_by\":\"auto\",\"created_at\":\"2026-04-26 15:46:35\",\"extension\":\"png\",\"order_by\":1,\"title\":\"Figure 1\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":205696,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eReceiver operating characteristic curves of the clinical, radiomics, and combined models based on repeated cross-validation predictions. The combined model achieved the highest point-estimate AUC.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"floatimage1.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-9427497/v1/7b3b84c83b09862a957faca6.png\"},{\"id\":107834646,\"identity\":\"bd0c0cc4-185f-4900-adf2-c41a6d726a77\",\"added_by\":\"auto\",\"created_at\":\"2026-04-26 15:46:35\",\"extension\":\"png\",\"order_by\":2,\"title\":\"Figure 2\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":246670,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eCalibration curves of the clinical, radiomics, and combined models. The combined model yielded the lowest Brier score, indicating the best overall agreement between predicted probabilities and observed outcomes.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"floatimage2.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-9427497/v1/3ba7d97eb4dc11b7660162b7.png\"},{\"id\":107834647,\"identity\":\"f9273ce6-b619-48cb-9505-a59c70ba992d\",\"added_by\":\"auto\",\"created_at\":\"2026-04-26 15:46:35\",\"extension\":\"png\",\"order_by\":3,\"title\":\"Figure 3\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":96236,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eBootstrap-based performance comparison across AUC, accuracy, sensitivity, specificity, and Brier score. Bars show point estimates and error bars indicate 95% confidence intervals.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"floatimage3.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-9427497/v1/af51b9d1a740109ed5577086.png\"},{\"id\":107834649,\"identity\":\"e594e496-d3be-46d3-bf9b-c8384fb8e91a\",\"added_by\":\"auto\",\"created_at\":\"2026-04-26 15:46:35\",\"extension\":\"png\",\"order_by\":4,\"title\":\"Figure 4\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":102699,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003ePairwise AUC differences estimated using 1000 bootstrap resamples. All confidence intervals crossed zero, indicating that no stable superiority of one model over another was established in this sample.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"floatimage4.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-9427497/v1/47ca944bae4a4c4bed1fe5e6.png\"},{\"id\":107870388,\"identity\":\"35422125-94dc-4518-803c-c93c0f832670\",\"added_by\":\"auto\",\"created_at\":\"2026-04-27 07:39:33\",\"extension\":\"png\",\"order_by\":5,\"title\":\"Figure 5\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":320275,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eMost frequently selected radiomics features across the 100 outer-fold training runs generated by 20 repeats of 5-fold stratified cross-validation. Features related to heterogeneity, size-zone distribution, and shape were repeatedly retained.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"floatimage5.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-9427497/v1/d64ccfb39db00c9394ea06dd.png\"},{\"id\":107874197,\"identity\":\"565fe474-22a9-4bfb-a359-361ed50af44e\",\"added_by\":\"auto\",\"created_at\":\"2026-04-27 08:05:45\",\"extension\":\"pdf\",\"order_by\":0,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"manuscript-pdf\",\"size\":1175260,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"manuscript.pdf\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-9427497/v1/fc3b3ede-de49-4954-87a9-0730dedd9ce2.pdf\"}],\"financialInterests\":\"No competing interests reported.\",\"formattedTitle\":\"Preoperative Prediction of Histological Grade in Pancreatic Neuroendocrine Tumors Using Whole-Pancreas CT Radiomics and Clinical Features: A Single-Center Retrospective Study\",\"fulltext\":[{\"header\":\"1. Introduction\",\"content\":\"\\u003cp\\u003ePancreatic neuroendocrine tumors (pNETs) are a heterogeneous group of neoplasms with markedly different biological behaviors and clinical outcomes[1]. Histological grade is one of the most important determinants of prognosis and treatment selection. In clinical practice, low-grade tumors may be considered for surveillance or limited surgery in selected circumstances, whereas higher-grade tumors generally require a more aggressive therapeutic strategy[2,3]. Accurate preoperative estimation of tumor grade is therefore central to personalized management.\\u003c/p\\u003e \\u003cp\\u003eAt present, preoperative grading relies mainly on endoscopic ultrasound-guided fine-needle aspiration or biopsy[4,5]. Although this approach remains important, it is invasive and may be affected by sampling error because of intratumoral heterogeneity[5]. Conventional contrast-enhanced CT can depict tumor size, location, enhancement pattern, and metastatic status, but visual assessment alone is often insufficient for reliable estimation of histological grade[6,7].\\u003c/p\\u003e \\u003cp\\u003eRadiomics provides a noninvasive approach for mining quantitative image information that may capture latent tumor heterogeneity[8]. Most previous studies focused on manually outlined tumor regions only. In contrast, whole-pancreas segmentation may additionally reflect subtle changes in the surrounding pancreatic parenchyma and tumor-associated microenvironment[9,10]. The present study therefore aimed to construct and compare three preoperative grading strategies based on whole-pancreas CT radiomics, clinical variables, and their combination, using a repeated nested cross-validation framework to reduce optimistic bias and to provide a more robust estimate of internal performance[11].\\u003c/p\\u003e\"},{\"header\":\"2. Materials and Methods\",\"content\":\"\\u003cdiv id=\\\"Sec3\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e2.1 Study design and patient population\\u003c/h2\\u003e \\u003cp\\u003eThis was a retrospective single-center study that included consecutive patients with pathologically confirmed pNETs treated at the First Affiliated Hospital of Fujian Medical University between January 2019 and December 2024. Eligible patients had available preoperative CT images and definitive postoperative pathological grading. Patients with incomplete imaging data, unavailable pathology, or prior treatment before CT were excluded.\\u003c/p\\u003e \\u003cp\\u003eThe primary study endpoint was histological grade recoded as a binary outcome: G1 tumors were defined as the low-grade group and G2/G3 tumors were combined as the high-grade group. No fixed training/testing split was used. Instead, all patients contributed to model development and internal validation through repeated nested cross-validation.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec4\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e2.2 CT imaging, whole-pancreas segmentation, and radiomics workflow\\u003c/h2\\u003e \\u003cp\\u003eCT examinations were acquired on multidetector scanners using routine institutional protocols. Whole-pancreas segmentation was performed manually using ITK-SNAP by a radiologist with 3 years of experience and reviewed by a senior radiologist with 20 years of experience. In a subset of randomly selected cases, repeated segmentation was used to assess reproducibility, and only stable features were retained for downstream analysis.\\u003c/p\\u003e \\u003cp\\u003eRadiomics features were extracted from multiphasic CT images using PyRadiomics (version 3.0). Within each training fold, features underwent z-score standardization, near-zero variance filtering, correlation-based redundancy reduction, and L1-penalized logistic feature selection. The selected features were combined into a fold-specific radiomics score (Rad-score)[12].\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec5\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e2.3 Clinical variable selection and model development\\u003c/h2\\u003e \\u003cp\\u003eCandidate clinical variables were screened using univariate analysis and clinical interpretability. Four variables were ultimately retained for the clinical model: Ki-67, maximum tumor diameter, T stage, and liver metastasis[13,14]. Three logistic regression models were then built: (1) a clinical model using the four clinical predictors, (2) a radiomics model using the fold-specific Rad-score, and (3) a combined model integrating the clinical variables with the Rad-score.\\u003c/p\\u003e \\u003cp\\u003eInternal validation used repeated stratified 5-fold cross-validation in the outer loop (20 repeats, yielding 100 outer-fold evaluations) and stratified 3-fold cross-validation in the inner loop for hyperparameter tuning and feature selection. This design ensured that all preprocessing and feature selection steps were repeated strictly within the training data to avoid information leakage.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec6\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e2.4 Model evaluation and statistical analysis\\u003c/h2\\u003e \\u003cp\\u003eModel discrimination and classification performance were summarized using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, F1 score, and Brier score. Out-of-fold predictions from repeated cross-validation were used to generate ROC and calibration plots. Bootstrap resampling (1000 iterations) was performed to estimate 95% confidence intervals for each performance metric and for pairwise differences in AUC.\\u003c/p\\u003e \\u003cp\\u003eContinuous variables are presented as median (interquartile range) and categorical variables as counts (percentages). Group comparisons were performed with the Mann-Whitney U test, chi-square test, or Fisher exact test as appropriate. All tests were two-sided, and P values\\u0026thinsp;\\u0026lt;\\u0026thinsp;0.05 were considered statistically significant.\\u003c/p\\u003e \\u003c/div\\u003e\"},{\"header\":\"3. Results\",\"content\":\"\\u003cdiv id=\\\"Sec8\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e3.1 Patient characteristics\\u003c/h2\\u003e \\u003cp\\u003eA total of 79 patients were included, including 31 with G1 tumors and 48 with G2/G3 tumors. Baseline comparisons between the two grade groups are summarized in Table\\u0026nbsp;\\u003cspan refid=\\\"Tab1\\\" class=\\\"InternalRef\\\"\\u003e1\\u003c/span\\u003e. Patients in the high-grade group had larger tumors (median 26.8 mm vs 19.0 mm, P\\u0026thinsp;=\\u0026thinsp;0.002), higher Ki-67 values (median 5.0% vs 1.0%, P\\u0026thinsp;\\u0026lt;\\u0026thinsp;0.001), and more advanced T stage (P\\u0026thinsp;=\\u0026thinsp;0.019). Age and liver metastasis did not differ significantly between the two groups.\\u003c/p\\u003e \\u003cp\\u003e \\u003cdiv class=\\\"gridtable\\\"\\u003e\\u003ctable float=\\\"Yes\\\" id=\\\"Tab1\\\" border=\\\"1\\\"\\u003e \\u003ccaption language=\\\"En\\\"\\u003e \\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 1\\u003c/div\\u003e \\u003cdiv class=\\\"CaptionContent\\\"\\u003e \\u003cp\\u003eBaseline characteristics according to histological grade group.\\u003c/p\\u003e \\u003c/div\\u003e \\u003c/caption\\u003e \\u003ccolgroup cols=\\\"4\\\"\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c1\\\" colnum=\\\"1\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c2\\\" colnum=\\\"2\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c3\\\" colnum=\\\"3\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c4\\\" colnum=\\\"4\\\"\\u003e\\u003c/div\\u003e \\u003cthead\\u003e \\u003ctr\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eVariable\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eG1 (n\\u0026thinsp;=\\u0026thinsp;31)\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eG2/G3 (n\\u0026thinsp;=\\u0026thinsp;48)\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eP value\\u003c/p\\u003e \\u003c/th\\u003e \\u003c/tr\\u003e \\u003c/thead\\u003e \\u003ctbody\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eAge at diagnosis, years\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e58.0 (54.0-69.5)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e57.5 (46.5\\u0026ndash;68.0)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.154\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eMaximum tumor diameter, mm\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e19.0 (15.0-25.3)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e26.8 (20.0-40.4)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.002\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eKi-67, %\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e1.0 (1.0\\u0026ndash;2.0)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e5.0 (3.4\\u0026ndash;13.9)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e\\u0026lt;\\u0026thinsp;0.001\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eT stage\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.019\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eT1\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e16 (51.6%)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e11 (22.9%)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eT2\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e12 (38.7%)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e24 (50.0%)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eT3\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e3 (9.7%)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e13 (27.1%)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eLiver metastasis, yes\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e3 (9.7%)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e12 (25.0%)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.141\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003c/tbody\\u003e \\u003c/colgroup\\u003e \\u003c/table\\u003e\\u003c/div\\u003e \\u003c/p\\u003e \\u003cp\\u003e \\u003cem\\u003eData are presented as median (interquartile range) or n (%). P values were derived using the Mann-Whitney U test, chi-square test, or Fisher exact test, as appropriate.\\u003c/em\\u003e \\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec9\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e3.2 Internal validation performance of the three models\\u003c/h2\\u003e \\u003cp\\u003eInternal validation performance is summarized in Table\\u0026nbsp;\\u003cspan refid=\\\"Tab2\\\" class=\\\"InternalRef\\\"\\u003e2\\u003c/span\\u003e. The combined model demonstrated the best overall balance of discrimination and calibration, with an AUC of 0.798, accuracy of 0.759, sensitivity of 0.771, specificity of 0.742, F1 score of 0.796, and Brier score of 0.182. The clinical model provided slightly higher specificity (0.806), whereas the radiomics-only model achieved the highest specificity (0.903) but lower sensitivity (0.563). ROC and calibration plots are shown in Figs.\\u0026nbsp;\\u003cspan refid=\\\"Fig1\\\" class=\\\"InternalRef\\\"\\u003e1\\u003c/span\\u003e and \\u003cspan refid=\\\"Fig2\\\" class=\\\"InternalRef\\\"\\u003e2\\u003c/span\\u003e, and the bootstrap-based comparison across metrics is shown in Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig3\\\" class=\\\"InternalRef\\\"\\u003e3\\u003c/span\\u003e.\\u003c/p\\u003e \\u003cp\\u003e \\u003cdiv class=\\\"gridtable\\\"\\u003e\\u003ctable float=\\\"Yes\\\" id=\\\"Tab2\\\" border=\\\"1\\\"\\u003e \\u003ccaption language=\\\"En\\\"\\u003e \\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 2\\u003c/div\\u003e \\u003cdiv class=\\\"CaptionContent\\\"\\u003e \\u003cp\\u003eBootstrap-based internal validation performance of the three prediction models.\\u003c/p\\u003e \\u003c/div\\u003e \\u003c/caption\\u003e \\u003ccolgroup cols=\\\"4\\\"\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c1\\\" colnum=\\\"1\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c2\\\" colnum=\\\"2\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c3\\\" colnum=\\\"3\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c4\\\" colnum=\\\"4\\\"\\u003e\\u003c/div\\u003e \\u003cthead\\u003e \\u003ctr\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eMetric\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eClinical model\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eRadiomics model\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eCombined model\\u003c/p\\u003e \\u003c/th\\u003e \\u003c/tr\\u003e \\u003c/thead\\u003e \\u003ctbody\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eAUC\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.780 (0.676\\u0026ndash;0.878)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.776 (0.660\\u0026ndash;0.881)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.798 (0.680\\u0026ndash;0.897)\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eAccuracy\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.696 (0.595\\u0026ndash;0.797)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.696 (0.595\\u0026ndash;0.797)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.759 (0.658\\u0026ndash;0.848)\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eSensitivity\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.625 (0.479\\u0026ndash;0.756)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.562 (0.413\\u0026ndash;0.714)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.771 (0.636\\u0026ndash;0.886)\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eSpecificity\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.806 (0.656\\u0026ndash;0.935)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.903 (0.783-1.000)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.742 (0.576\\u0026ndash;0.893)\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eF1 score\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.714 (0.585\\u0026ndash;0.813)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.692 (0.557\\u0026ndash;0.804)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.796 (0.690\\u0026ndash;0.880)\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eBrier score\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.214 (0.188\\u0026ndash;0.242)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.237 (0.228\\u0026ndash;0.246)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.182 (0.130\\u0026ndash;0.247)\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003c/tbody\\u003e \\u003c/colgroup\\u003e \\u003c/table\\u003e\\u003c/div\\u003e \\u003c/p\\u003e \\u003cp\\u003e \\u003cem\\u003eValues are point estimates with bootstrap 95% confidence intervals in parentheses. The combined model had the lowest Brier score, indicating the best overall calibration.\\u003c/em\\u003e \\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec10\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e3.3 Pairwise AUC comparison and radiomics feature stability\\u003c/h2\\u003e \\u003cp\\u003eDespite the favorable point estimates of the combined model, bootstrap analysis of pairwise AUC differences did not demonstrate statistically stable superiority (Table\\u0026nbsp;\\u003cspan refid=\\\"Tab3\\\" class=\\\"InternalRef\\\"\\u003e3\\u003c/span\\u003e and Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig4\\\" class=\\\"InternalRef\\\"\\u003e4\\u003c/span\\u003e). The delta AUC was 0.015 for combined minus clinical and 0.021 for combined minus radiomics, and both confidence intervals crossed zero. Repeated training runs showed that radiomics features related to size-zone non-uniformity, dependence variance, size, and first-order intensity were most frequently selected (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig5\\\" class=\\\"InternalRef\\\"\\u003e5\\u003c/span\\u003e). In the final combined model fitted on the full dataset, the coefficient of the Rad-score (9.640) was larger than those of the individual clinical variables, suggesting that the aggregated radiomics signature remained an important contributor to prediction.\\u003c/p\\u003e \\u003cp\\u003e \\u003cdiv class=\\\"gridtable\\\"\\u003e\\u003ctable float=\\\"Yes\\\" id=\\\"Tab3\\\" border=\\\"1\\\"\\u003e \\u003ccaption language=\\\"En\\\"\\u003e \\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 3\\u003c/div\\u003e \\u003cdiv class=\\\"CaptionContent\\\"\\u003e \\u003cp\\u003ePairwise differences in AUC estimated by 1000 bootstrap resamples.\\u003c/p\\u003e \\u003c/div\\u003e \\u003c/caption\\u003e \\u003ccolgroup cols=\\\"3\\\"\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c1\\\" colnum=\\\"1\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c2\\\" colnum=\\\"2\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c3\\\" colnum=\\\"3\\\"\\u003e\\u003c/div\\u003e \\u003cthead\\u003e \\u003ctr\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eComparison\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eDelta AUC\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e95% CI\\u003c/p\\u003e \\u003c/th\\u003e \\u003c/tr\\u003e \\u003c/thead\\u003e \\u003ctbody\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eCombined minus clinical\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.015\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e-0.116 to 0.138\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eCombined minus radiomics\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.021\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e-0.037 to 0.065\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eClinical minus radiomics\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.006\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e-0.133 to 0.155\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003c/tbody\\u003e \\u003c/colgroup\\u003e \\u003c/table\\u003e\\u003c/div\\u003e \\u003c/p\\u003e \\u003cp\\u003e \\u003cem\\u003eConfidence intervals crossing zero indicate that a stable difference in discrimination was not established.\\u003c/em\\u003e \\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003c/div\\u003e\"},{\"header\":\"4. Discussion\",\"content\":\"\\u003cp\\u003eThis study developed a whole-pancreas CT radiomics strategy for preoperative prediction of pNET histological grade and compared it with a clinical-only model and a combined model. The main finding was that the combined model showed the most balanced internal validation profile, with the highest AUC and F1 score as well as the lowest Brier score. However, the estimated performance gain over the single-domain models was modest, and bootstrap confidence intervals for the AUC differences crossed zero. The results therefore support cautious interpretation rather than a claim of clear superiority[15,16].\\u003c/p\\u003e \\u003cp\\u003eAn important feature of the present work is the use of whole-pancreas segmentation rather than tumor-only segmentation. Whole-organ analysis may capture subtle alterations in the pancreatic background parenchyma, vascular environment, or tumor-associated heterogeneity that are not fully represented by a focal tumor ROI[17]. The frequent selection of radiomics features associated with heterogeneity and size-zone distribution also supports the biological plausibility of the radiomics signature[18,19].\\u003c/p\\u003e \\u003cp\\u003eThe current modeling strategy was intentionally conservative. Instead of relying on a single random training/testing split, we used repeated nested stratified cross-validation together with bootstrap-based uncertainty estimation[20]. This approach is better suited to small datasets and helps reduce optimistic bias. The trade-off is that any apparent advantage of one model over another must remain large and stable to survive resampling-based comparison[21,22]. In the present cohort, the combined model appeared promising but not definitively superior.\\u003c/p\\u003e \\u003cp\\u003eSeveral limitations should be acknowledged. First, this was a single-center retrospective study with a relatively small sample size. Second, no external validation cohort was available, and therefore generalizability remains uncertain. Third, although the combined model incorporated key clinical variables and radiomics information, other potentially relevant biomarkers were not evaluated in the current framework. Future multicenter studies should test the transportability of the model, refine feature harmonization across scanners, and explore integration with additional imaging or molecular markers.\\u003c/p\\u003e\"},{\"header\":\"5. Conclusion\",\"content\":\"\\u003cp\\u003e Whole-pancreas CT radiomics combined with clinical variables provided the best overall internal validation performance for the preoperative prediction of pNET histological grade in this cohort. Nevertheless, the incremental discrimination gain over the clinical and radiomics-only models was small and not statistically stable in bootstrap comparison. External validation in larger cohorts is required before clinical implementation.\\u003c/p\\u003e\"},{\"header\":\"Declarations\",\"content\":\"\\u003cp\\u003e \\u003cstrong\\u003eEthics approval:\\u003c/strong\\u003e \\u003cp\\u003e Ethical approval (2023J0112) was obtained from the ethics review board of The Fujian Medical University; the requirement for informed consent was waived owing to the retrospective nature of the study.\\u003c/p\\u003e \\u003c/p\\u003e\\u003cp\\u003e \\u003ch2\\u003eConflict of interest:\\u003c/h2\\u003e \\u003cp\\u003eThe authors declare no competing interests.\\u003c/p\\u003e \\u003c/p\\u003e\\u003ch2\\u003eFunding:\\u003c/h2\\u003e \\u003cp\\u003eThis work was supported by 1. Fujian Provincial Natural Science Foundation (2023J0112); 2. Fujian Provincial Financial Special Fund (BRB-2024WRX) 3. Wu Jieping Medical Foundation(320.6750.2023-05-116).\\u003c/p\\u003e\\u003ch2\\u003eAuthor Contribution\\u003c/h2\\u003e\\u003cp\\u003eAuthor contributionsXiming Liu contributed to study conception and design, collected the clinical and imaging data, and drafted the manuscript. Yujie Zhou contributed to data analysis, model development, and manuscript revision. Shuang Gao contributed to statistical analysis and revision of the manuscript. Wenpei Zhang contributed to data curation, methodological discussion, and manuscript editing. Rixiong Wang conceived and supervised the study, interpreted the results, and critically revised the manuscript for important intellectual content. All authors reviewed and approved the final manuscript.\\u003c/p\\u003e\"},{\"header\":\"Unsectioned Paragraphs\",\"content\":\"\\u003col\\u003e\\n\\u003cli\\u003eMa, Z.-Y. \\u003cem\\u003eet al.\\u003c/em\\u003e Pancreatic neuroendocrine tumors: A review of serum biomarkers, staging, and management. \\u003cem\\u003eWorld J. Gastroenterol.\\u003c/em\\u003e \\u003cstrong\\u003e26\\u003c/strong\\u003e, 2305\\u0026ndash;2322 (2020).\\u003c/li\\u003e\\n\\u003cli\\u003eN F. \\u003cem\\u003eet al.\\u003c/em\\u003e Natural history and clinical outcomes of pancreatic neuroendocrine neoplasms based on the WHO 2017 classification; a single-center experience of 30 years. \\u003cem\\u003ePancreatol. Off. J. Int. Assoc. Pancreatol. IAP Al\\u003c/em\\u003e \\u003cstrong\\u003e20\\u003c/strong\\u003e, (2020).\\u003c/li\\u003e\\n\\u003cli\\u003ePavel, M. \\u003cem\\u003eet al.\\u003c/em\\u003e Gastroenteropancreatic neuroendocrine neoplasms: ESMO Clinical Practice Guidelines for diagnosis, treatment and follow-up. \\u003cem\\u003eAnn. Oncol. Off. J. Eur. Soc. Med. Oncol.\\u003c/em\\u003e \\u003cstrong\\u003e31\\u003c/strong\\u003e, 844\\u0026ndash;860 (2020).\\u003c/li\\u003e\\n\\u003cli\\u003eArdengh, J. C. \\u003cem\\u003eet al.\\u003c/em\\u003e Pancreatic splenosis mimicking neuroendocrine tumors: microhistological diagnosis by endoscopic ultrasound guided fine needle aspiration. \\u003cem\\u003eArq. Gastroenterol.\\u003c/em\\u003e \\u003cstrong\\u003e50\\u003c/strong\\u003e, 10\\u0026ndash;14 (2013).\\u003c/li\\u003e\\n\\u003cli\\u003eYe, X. \\u003cem\\u003eet al.\\u003c/em\\u003e Endoscopic Ultrasound-Guided Fine Needle Acquisition for Evaluation of Pancreatic Neuroendocrine Tumors: A Meta-Analysis. \\u003cem\\u003eJ. Clin. Gastroenterol.\\u003c/em\\u003e \\u003cstrong\\u003e59\\u003c/strong\\u003e, 310\\u0026ndash;320 (2025).\\u003c/li\\u003e\\n\\u003cli\\u003eLee, L., Ito, T. \\u0026amp; Jensen, R. T. Imaging of pancreatic neuroendocrine tumors: recent advances, current status, and controversies. \\u003cem\\u003eExpert Rev. Anticancer Ther.\\u003c/em\\u003e \\u003cstrong\\u003e18\\u003c/strong\\u003e, 837\\u0026ndash;860 (2018).\\u003c/li\\u003e\\n\\u003cli\\u003eTacelli, M. \\u003cem\\u003eet al.\\u003c/em\\u003e Reliability of grading preoperative pancreatic neuroendocrine tumors on EUS specimens: a systematic review with meta-analysis of aggregate and individual data. \\u003cem\\u003eGastrointest. Endosc.\\u003c/em\\u003e \\u003cstrong\\u003e96\\u003c/strong\\u003e, 898-908.e23 (2022).\\u003c/li\\u003e\\n\\u003cli\\u003eLambin, P. \\u003cem\\u003eet al.\\u003c/em\\u003e Radiomics: the bridge between medical imaging and personalized medicine. \\u003cem\\u003eNat. Rev. Clin. Oncol.\\u003c/em\\u003e \\u003cstrong\\u003e14\\u003c/strong\\u003e, 749\\u0026ndash;762 (2017).\\u003c/li\\u003e\\n\\u003cli\\u003eKumar, V. \\u003cem\\u003eet al.\\u003c/em\\u003e Radiomics: the process and the challenges. \\u003cem\\u003eMagn. Reson. Imaging\\u003c/em\\u003e \\u003cstrong\\u003e30\\u003c/strong\\u003e, 1234\\u0026ndash;1248 (2012).\\u003c/li\\u003e\\n\\u003cli\\u003eWang, S. \\u003cem\\u003eet al.\\u003c/em\\u003e Compute Tomography Radiomics Analysis on Whole Pancreas Between Healthy Individual and Pancreatic Ductal Adenocarcinoma Patients: Uncertainty Analysis and Predictive Modeling. \\u003cem\\u003eTechnol. Cancer Res. Treat.\\u003c/em\\u003e \\u003cstrong\\u003e21\\u003c/strong\\u003e, 15330338221126869 (2022).\\u003c/li\\u003e\\n\\u003cli\\u003eW G. \\u003cem\\u003eet al.\\u003c/em\\u003e Development and validation of CT-based radiomics deep learning signatures to predict lymph node metastasis in non-functional pancreatic neuroendocrine tumors: a multicohort study. \\u003cem\\u003eEClinicalMedicine\\u003c/em\\u003e \\u003cstrong\\u003e65\\u003c/strong\\u003e, (2023).\\u003c/li\\u003e\\n\\u003cli\\u003eShi, Z., Traverso, A., van Soest, J., Dekker, A. \\u0026amp; Wee, L. Technical Note: Ontology-guided radiomics analysis workflow (O-RAW). \\u003cem\\u003eMed. Phys.\\u003c/em\\u003e \\u003cstrong\\u003e46\\u003c/strong\\u003e, 5677\\u0026ndash;5684 (2019).\\u003c/li\\u003e\\n\\u003cli\\u003eF, G. \\u003cem\\u003eet al.\\u003c/em\\u003e KI-67 heterogeneity in well differentiated gastro-entero-pancreatic neuroendocrine tumors: when is biopsy reliable for grade assessment? \\u003cem\\u003eEndocrine\\u003c/em\\u003e \\u003cstrong\\u003e57\\u003c/strong\\u003e, (2017).\\u003c/li\\u003e\\n\\u003cli\\u003eDevelopment and validation of a prognostic nomogram based on inflammation and Nutrition-Related Indexes for predicting postoperative survival in patients with pancreatic neuroendocrine tumors - PubMed. https://pubmed.ncbi.nlm.nih.gov/41469599/.\\u003c/li\\u003e\\n\\u003cli\\u003eSnell, K. I. E. \\u003cem\\u003eet al.\\u003c/em\\u003e Transparent reporting of multivariable prediction models for individual prognosis or diagnosis: checklist for systematic reviews and meta-analyses (TRIPOD-SRMA). \\u003cem\\u003eBMJ\\u003c/em\\u003e \\u003cstrong\\u003e381\\u003c/strong\\u003e, e073538 (2023).\\u003c/li\\u003e\\n\\u003cli\\u003eZwanenburg, A. \\u003cem\\u003eet al.\\u003c/em\\u003e The Image Biomarker Standardization Initiative: Standardized Quantitative Radiomics for High-Throughput Image-based Phenotyping. \\u003cem\\u003eRadiology\\u003c/em\\u003e \\u003cstrong\\u003e295\\u003c/strong\\u003e, 328\\u0026ndash;338 (2020).\\u003c/li\\u003e\\n\\u003cli\\u003eKang, W. \\u003cem\\u003eet al.\\u003c/em\\u003e Application of radiomics-based multiomics combinations in the tumor microenvironment and cancer prognosis. \\u003cem\\u003eJ. Transl. Med.\\u003c/em\\u003e \\u003cstrong\\u003e21\\u003c/strong\\u003e, 598 (2023).\\u003c/li\\u003e\\n\\u003cli\\u003eLambin, P. \\u003cem\\u003eet al.\\u003c/em\\u003e Radiomics: extracting more information from medical images using advanced feature analysis. \\u003cem\\u003eEur. J. Cancer\\u003c/em\\u003e \\u003cstrong\\u003e48\\u003c/strong\\u003e, 441\\u0026ndash;446 (2012).\\u003c/li\\u003e\\n\\u003cli\\u003eDou, T. H., Coroller, T. P., van Griethuysen, J. J. M., Mak, R. H. \\u0026amp; Aerts, H. J. W. L. Peritumoral radiomics features predict distant metastasis in locally advanced NSCLC. \\u003cem\\u003ePloS One\\u003c/em\\u003e \\u003cstrong\\u003e13\\u003c/strong\\u003e, e0206108 (2018).\\u003c/li\\u003e\\n\\u003cli\\u003eSchroeder, A. W., Tran, Z., Sexton, K. \\u0026amp; Salzberg, A. D. Clinician\\u0026rsquo;s Guide to Artificial Intelligence: Technical Foundations of Machine Learning. \\u003cem\\u003eMed. Clin. North Am.\\u003c/em\\u003e \\u003cstrong\\u003e110\\u003c/strong\\u003e, 287\\u0026ndash;305 (2026).\\u003c/li\\u003e\\n\\u003cli\\u003eTRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods - PubMed. https://pubmed.ncbi.nlm.nih.gov/38626948/.\\u003c/li\\u003e\\n\\u003cli\\u003eSnell, K. I. E. \\u003cem\\u003eet al.\\u003c/em\\u003e Transparent reporting of multivariable prediction models for individual prognosis or diagnosis: checklist for systematic reviews and meta-analyses (TRIPOD-SRMA). \\u003cem\\u003eBMJ\\u003c/em\\u003e\\u003cstrong\\u003e381\\u003c/strong\\u003e, e073538 (2023).\\u003c/li\\u003e\\n\\u003c/ol\\u003e\"}],\"fulltextSource\":\"\",\"fullText\":\"\",\"funders\":[],\"hasAdminPriorityOnWorkflow\":false,\"hasManuscriptDocX\":true,\"hasOptedInToPreprint\":true,\"hasPassedJournalQc\":\"\",\"hasAnyPriority\":false,\"hideJournal\":false,\"highlight\":\"\",\"institution\":\"\",\"isAcceptedByJournal\":false,\"isAuthorSuppliedPdf\":false,\"isDeskRejected\":\"\",\"isHiddenFromSearch\":false,\"isInQc\":false,\"isInWorkflow\":false,\"isPdf\":false,\"isPdfUpToDate\":true,\"isWithdrawnOrRetracted\":false,\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"abdominal-radiology\",\"isNatureJournal\":false,\"hasQc\":true,\"allowDirectSubmit\":false,\"externalIdentity\":\"aima\",\"sideBox\":\"Learn more about [Abdominal Radiology](http://link.springer.com/journal/261)\",\"snPcode\":\"261\",\"submissionUrl\":\"https://submission.springernature.com/new-submission/261/3\",\"title\":\"Abdominal Radiology\",\"twitterHandle\":\"\",\"acdcEnabled\":true,\"dfaEnabled\":true,\"editorialSystem\":\"stoa\",\"reportingPortfolio\":\"Springer Hybrid\",\"inReviewEnabled\":true,\"inReviewRevisionsEnabled\":false},\"keywords\":\"pancreatic neuroendocrine tumors, CT radiomics, whole-pancreas segmentation, histological grade, combined model, bootstrap validation\",\"lastPublishedDoi\":\"10.21203/rs.3.rs-9427497/v1\",\"lastPublishedDoiUrl\":\"https://doi.org/10.21203/rs.3.rs-9427497/v1\",\"license\":{\"name\":\"CC BY 4.0\",\"url\":\"https://creativecommons.org/licenses/by/4.0/\"},\"manuscriptAbstract\":\"\\u003ch2\\u003eObjective\\u003c/h2\\u003e \\u003cp\\u003eTo develop and internally validate a whole-pancreas CT radiomics model, a clinical model, and a combined model for the preoperative prediction of histological grade in pancreatic neuroendocrine tumors (pNETs).\\u003c/p\\u003e\\u003ch2\\u003eMethods\\u003c/h2\\u003e \\u003cp\\u003eThis single-center retrospective study included 79 pathologically confirmed pNETs between January 2019 and December 2024. The endpoint was binary histological grade (G1 vs G2/G3). Whole-pancreas segmentation was performed on preoperative multiphasic CT images, and radiomics signatures were generated within each training fold using standardization, near-zero variance filtering, redundancy reduction, and L1-penalized feature selection. Four clinical variables (Ki-67, maximum tumor diameter, T stage, and liver metastasis) were selected based on univariate analysis and clinical relevance. Three logistic models were developed: a clinical model, a radiomics model, and a combined model. Internal validation used nested repeated stratified 5-fold cross-validation (20 repeats in the outer loop and 3 folds in the inner loop) with 1000 bootstrap resamples to estimate 95% confidence intervals (CIs) for performance metrics.\\u003c/p\\u003e\\u003ch2\\u003eResults\\u003c/h2\\u003e \\u003cp\\u003eOf the 79 patients, 31 had G1 tumors and 48 had G2/G3 tumors. The combined model showed the best overall performance, with an AUC of 0.798, accuracy of 0.759, sensitivity of 0.771, specificity of 0.742, F1 score of 0.796, and Brier score of 0.182. The clinical and radiomics models yielded AUCs of 0.780 and 0.776, respectively. Bootstrap comparison showed no statistically stable AUC difference between the combined and clinical models (delta AUC 0.015, 95% CI -0.116 to 0.138) or between the combined and radiomics models (delta AUC 0.021, 95% CI -0.037 to 0.065).\\u003c/p\\u003e\\u003ch2\\u003eConclusion\\u003c/h2\\u003e \\u003cp\\u003eA whole-pancreas CT radiomics-clinical combined model showed the most balanced internal validation performance for preoperative grading of pNETs, although its incremental benefit over the single-domain models remains uncertain in this small cohort. Larger multicenter studies with external validation are needed.\\u003c/p\\u003e\",\"manuscriptTitle\":\"Preoperative Prediction of Histological Grade in Pancreatic Neuroendocrine Tumors Using Whole-Pancreas CT Radiomics and Clinical Features: A Single-Center Retrospective Study\",\"msid\":\"\",\"msnumber\":\"\",\"nonDraftVersions\":[{\"code\":1,\"date\":\"2026-04-26 15:46:30\",\"doi\":\"10.21203/rs.3.rs-9427497/v1\",\"editorialEvents\":[{\"type\":\"communityComments\",\"content\":0},{\"type\":\"reviewerAgreed\",\"content\":\"18534646540441207219511653460781423278\",\"date\":\"2026-05-07T07:51:36+00:00\",\"index\":\"hide\",\"fulltext\":\"\"},{\"type\":\"reviewerAgreed\",\"content\":\"299830754739986319491211558194063150140\",\"date\":\"2026-05-06T08:05:46+00:00\",\"index\":\"hide\",\"fulltext\":\"\"},{\"type\":\"reviewersInvited\",\"content\":\"\",\"date\":\"2026-04-17T00:35:20+00:00\",\"index\":\"\",\"fulltext\":\"\"},{\"type\":\"editorAssigned\",\"content\":\"\",\"date\":\"2026-04-16T09:40:58+00:00\",\"index\":\"\",\"fulltext\":\"\"},{\"type\":\"checksComplete\",\"content\":\"\",\"date\":\"2026-04-16T09:40:02+00:00\",\"index\":\"\",\"fulltext\":\"\"},{\"type\":\"submitted\",\"content\":\"Abdominal Radiology\",\"date\":\"2026-04-15T12:55:00+00:00\",\"index\":\"\",\"fulltext\":\"\"}],\"status\":\"published\",\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"abdominal-radiology\",\"isNatureJournal\":false,\"hasQc\":true,\"allowDirectSubmit\":false,\"externalIdentity\":\"aima\",\"sideBox\":\"Learn more about [Abdominal Radiology](http://link.springer.com/journal/261)\",\"snPcode\":\"261\",\"submissionUrl\":\"https://submission.springernature.com/new-submission/261/3\",\"title\":\"Abdominal Radiology\",\"twitterHandle\":\"\",\"acdcEnabled\":true,\"dfaEnabled\":true,\"editorialSystem\":\"stoa\",\"reportingPortfolio\":\"Springer Hybrid\",\"inReviewEnabled\":true,\"inReviewRevisionsEnabled\":false}}],\"origin\":\"\",\"ownerIdentity\":\"e8296c25-904b-4c86-80c5-eef19e488d6d\",\"owner\":[],\"postedDate\":\"April 26th, 2026\",\"published\":true,\"recentEditorialEvents\":[{\"type\":\"reviewerAgreed\",\"content\":\"18534646540441207219511653460781423278\",\"date\":\"2026-05-07T07:51:36+00:00\",\"index\":29,\"fulltext\":\"\"},{\"type\":\"reviewerAgreed\",\"content\":\"299830754739986319491211558194063150140\",\"date\":\"2026-05-06T08:05:46+00:00\",\"index\":24,\"fulltext\":\"\"}],\"rejectedJournal\":[],\"revision\":\"\",\"amendment\":\"\",\"status\":\"under-review\",\"subjectAreas\":[],\"tags\":[],\"updatedAt\":\"2026-04-26T15:46:31+00:00\",\"versionOfRecord\":[],\"versionCreatedAt\":\"2026-04-26 15:46:30\",\"video\":\"\",\"vorDoi\":\"\",\"vorDoiUrl\":\"\",\"workflowStages\":[]},\"version\":\"v1\",\"identity\":\"rs-9427497\",\"journalConfig\":\"researchsquare\"},\"__N_SSP\":true},\"page\":\"/article/[identity]/[[...version]]\",\"query\":{\"redirect\":\"/article/rs-9427497\",\"identity\":\"rs-9427497\",\"version\":[\"v1\"]},\"buildId\":\"XKTyCvWXoU3ODBz1xrDgd\",\"isFallback\":false,\"isExperimentalCompile\":false,\"dynamicIds\":[84888],\"gssp\":true,\"scriptLoader\":[]}","source_license":"CC-BY-4.0","license_restricted":false}